基于奇异值分解的电感式磨粒传感器信号降噪方法研究

Study on the Denoising Method for Signals of Inductive Wear Debris Sensor Based on Singular Value Decomposition

利用奇异值降噪技术对含有系统噪声的电感式磨粒传感器测试信号进行处理。针对奇异值降噪中轨道矩阵最优重构阶次难以有效确定的问题,提出一种基于噪声阈值和奇异值聚类的重构阶次确定方法。首先由含噪信号轨道矩阵的嵌入维数和噪声方差确定奇异值的噪声阈值,得到重构阶次的上限;然后依据信号奇异值分布离散,噪声奇异值分布集中的特点,对大于噪声阈值的奇异值进行聚类分析,进一步确定轨道矩阵的重构阶次。仿真和实测信号降噪效果表明,该降噪算法能显著改善含噪信号的信噪比,降噪后的信号具有较小的峰值误差,适合电感式磨粒传感器信号的降噪。

By means of the denoising technology of singular value analysis,the signal of inductive wear debris sensor corrupted by system noise is processed.The optical reconstruction order of the trajectory matrix is difficult to determine for noise reduction in singular value decomposition.A new method is proposed to solve this problem based on noise threshold and singular values cluster.By the embedding dimension of the trajectory matrix and noise variance,the noise threshold is determined and the upper limit of order reconstruction is got at first.Then,according to the distribution characteristics of singular values that signal＇s singular values are disperse and noise＇s singular values are concentrated,the clustering analysis is done for the singular values that are greater than noise threshold.The final reconstruction order is determined based on the clustering results.The denoising results of simulation signals and testing signals show that this denoising method can improve SNR notably and the signals after noise reduction have less peak value errors.The method is suitable for the noise reduction of inductive wear debris signals.